@ariadne In practice none of this actually applies and pissnet is better coordinated than IRCNow. There’s no single staff channel (equivalent of pissnet’s #opers), and NgIRCd is horribly broken in ways too numerous to list (but the founder, jrmu, refuses to give up on NgIRCd and suggests that if you don’t like it you should start another network with another ircd and somehow bridge it).

Here’s a far-from-complete list of the ways in which NgIRCd is broken:

• There’s a bug that allows for s2s injection if charconv is enabled. This is now disabled on all IRCNow servers, but it took a few weeks to get everyone to do this. One server refused (claiming that disabling charconv would introduce other bugs somehow) and eventually got delinked and switched to unrealircd, but now seems to be back again.
• Sometimes random modes and bans get set on channels on netjoins, and there’s no way to unset them without a pseudoserver. I’m sure there’s a memory safety bug here somewhere.
• There’s no timestamps so it’s easy to abuse a netsplit to nick-collide someone off the network or take over a channel.
• On netjoins, servers send incorrect messages like MODE +o nick1 nick2 nick3 nick4, resulting in desyncs.
• Cloaking is horribly broken. There’s two kinds of random cloaks: one which is always enabled and one which can be enabled by setting umode +x. The configuration of this isn’t consistent across the network (some have no cloaking, some have umode +x cloaking, and some have always enabled cloaking), and servers apply umode +x cloaks remotely, which results in bans and G-lines not working properly access the network. I’m also not sure if there’s even a way to set CIDR bans, since K/G-lines apply on the cloaked IP rather than the real IP.
• Every K/G-line set leaks the oper’s real uncloaked IP address to the entire network (as well as the ban reason and mask the ban was set on), because anyone can set umode +s (or join &SERVER) to receive snotes.
• There’s no oper override logging, or way to opt out of oper override without completely disabling it for the server.
• It’s not modular and there’s no reloading the code without restarting the server.

It is literally the worst ircd I have ever used.

@stefano happy birthday!

@mapache system packages or binaries I compile myself if unavailable

@ariadne I didn't know what this was and actually thought you meant insurance for umbrellas (like in case they get damaged by wind) until I looked it up

@cwg1231 @alexia

The minio thing is shitty, but also not at all the same as what the OQL is suggesting. Also, the AGPL seems like it forbids selling license exceptions.

Correct. in order to sell exceptions you need to use a separate license for that. It’s kind-of like dual-licensing except instead of the alternative terms being available to anyone, they’re only available to those who buy them. The FSF wrote an article about this. I am using the terms “selling exceptions” and “selling an alternative license” interchangeably.

If you are the only copyright holder you can easily do that (no license can forbid the copyright from also releasing their code under another license), but if there are any other copyright holders then they must agree too, or you need some sort of CLA which gives the maintainer extra rights.

It’s exactly the same if you use the OQL instead of the AGPL. If you release your software under any license and accept contributions without some sort of copyright assignment or CLA (meaning you are not the sole copyright holder and have equal rights to the code as any other user), you cannot sell exceptions to that license.

That’s the relevance of the minio thing - I wasn’t saying it was similar to the OQL, I was saying that it’s a danger of using a CLA, which is necessary in order to sell license exceptions (whether to the OQL or the AGPL, or any other license) unless you are the only copyright holder. I’m not sure if there’s a way to mitigate this with a more restricted CLA.

The point isn’t scaring off companies, I want my software to be used, but only to better the world. The point is that under the current state of capitalism, incorporating OSS into your supply chain should require compensating developers commensurate to the responsibility you’re placing on them.

If I understand your point correctly, what you want is for companies that use your software to pay you. There are two ways to achieve this:

• Explicitly forbid them from using your software (using the OQL), and sell an alternative license.
• Scare them off from using your software without explicitly forbidding them (using the AGPL), and sell an alternative license.

They both achieve the same goal (companies will pay you for the alternative license), but the latter does so without making your software non-free. Perhaps the OQL is more effective at getting companies to pay (I’m not sure by how much), because some companies are fine with the AGPL, but using the AGPL and selling exceptions is a business model that people have successfully used.

@catsalad Three semicolons.

#include <stdio.h>
int main(void) {
;;;printf("Like this\n")
;;;return 0;
}

@dpk 2000 (Transmeta Crusoe). Also arguably P6 even before that (instructions were internally translated into RISC-y micro-operations)

@ariadne @ignaloidas Should I try writing a patch (and sending it to LKML) that adds vm.overcommit_memory=3 which would disable overcommit (as mode 2) but not ignore MAP_NORESERVE or has someone already done that and had it rejected for some reason?

@kemona_halftau It seems to be different every time it's requested rather than different per instance (I don't have a media proxy so it's getting requested from the original server every time)

@ariadne @ignaloidas MAP_NORESERVE is ignored if vm.overcommit_memory=2: https://docs.kernel.org/mm/overcommit-accounting.html. Unfortunately I don't think there's a way to opt in to overcommit for specific allocations if it is disabled.

fox (different per instance):

@ariadne @ska @navi @dakkar How do I actually achieve this using netfilter or eBPF?

nftables's skuid/skgid matching only works on output rules: I can easily prevent specific users from connecting to certain ports, but I'm not sure how this can be used to prevent specific users from listening on certain ports.

eBPF I don't know that much about but I'd be interested in how I can do this with it.

game over!
score: 98
││
││
││
││
││
││
││
││
││
││

I hope all those devs who went “I don’t need to optimize, there’s plenty of RAM” are going to get slapped with complaints about bad performance because the software runs like shit on new computers whose RAM size traveled back to 2015

@catsalad @binsk @us3r1d @alice Unfortunately not. I can view the text of the post but still not the image: lgbtqia.space geoblocks the UK and my method of bypassing it (specific routes for the IP address through a VPN interface) works for lgbtqia.space itself but not the CDN used for images (which uses GeoDNS so I'd need to use a DNS server hosted elsewhere to bypass it). I plan to set that up but I haven't done so yet.

@catsalad @us3r1d @binsk I clicked on it but that domain resolves to 127.0.0.1 (I'm GeoDNS-blocked)

@dpk I just tested now with mpv --term-status-msg='${audio-bitrate}' proxying via a server in Germany and it’s definitely 320 kbps

@dpk @mossfet

there was never access to 320 kbps feeds outside the UK

This isn’t true. When I was in Norway in July I was using the 320kbps worldwide HLS streams. I use this script for listening to BBC radio stations: https://paste.sr.ht/~noisytoot/baadc025381a8083b7d01822a7d58b8c3e465913